Reference-to-video (R2V) is a generative AI technique that uses one or more reference images to guide video generation — not as the literal first frame, but as an identity anchor. The model studies your references (a character's face, a product's design, an art style) and generates new scenes where that identity stays consistent.
Reference vs. first frame
This is the key distinction from image-to-video:
- Image-to-video treats your image as the opening frame. The video starts exactly there.
- Reference-to-video treats your images as a definition of what things look like. The model can then place that character or product into an entirely new scene, angle, or action described by your prompt.
In short: I2V continues a picture; R2V casts it.
Why it matters
Consistency is the hardest problem in AI video. Generate the same "red-haired girl in a yellow raincoat" twice from text alone and you'll get two different girls. Reference-to-video solves this:
- Character consistency — keep the same protagonist across shots, scenes, and episodes.
- Product fidelity — show your actual product from new angles, in new environments.
- Style continuity — carry an illustration style or brand look through a whole series.
Prompting tips
Give the model clean, well-lit references that show the subject clearly. Then let the prompt do the directing: new setting, new action, new camera. Say what should change — the references already say what should stay. More in our prompt writing guide.
Reference-to-video on Molyin
Molyin's Seedance 2.0 generator supports reference-to-video alongside text- and image-to-video: 5 or 10 second clips, up to 1080p, six aspect ratios, optional audio, and seed control.
Related terms
- Text-to-Video — generate from a written prompt alone.
- Image-to-Video — animate an image as the literal first frame.
Try it with the AI video generator.
